Page 1
Title: Social Networks in Emergency Response
Name: Dashun Wang1,2, Yu-Ru Lin3,4, James P. Bagrow5,6
Affil./Addr. 1: Center for Complex Network Research,
Northeastern University,
Boston, MA 02115, USA
E-mail: [email protected]
Affil./Addr. 2: Center for Cancer Systems Biology,
Dana-Farber Cancer Institute,
Boston, MA 02115, USA.
Affil./Addr. 3: College of Computer and Information Science,
Northeastern University,
Boston, MA, 02115, USA
E-mail: [email protected]
Affil./Addr. 4: Institute for Quantitative Social Science,
Harvard University,
Cambridge, MA 02138, USA
Affil./Addr. 5: Engineering Sciences and Applied Mathematics,
Northwestern University,
Evanston, IL 60208, USA
E-mail: [email protected]
Affil./Addr. 6: Northwestern Institute on Complex Systems,
Northwestern University,
Evanston, IL 60208, USA
Social Networks in Emergency
Response
Page 2
2
Synonyms
emergencies, disaster, collective response, social networks, communication networks,
event detection, spatiotemporal analysis, data mining
Glossary
Emergency: an unexpected and often dangerous situation, typically affecting multiple
individuals and requiring immediate action. Social and communication networks:
networks of people interacting with each other through web-based (e.g. Twitter) and
mobile based (e.g. mobile phone) technologies. Social media: Web-based tools that
enable people to communicate and interact with each other in various media forms
including text and multimedia. Examples of these tools include emails, instant mes-
sengers (IM), blogs, micro blogs (e.g. Twitter), vlogs (e.g. YouTube), podcasts, forum,
wikis, social news (e.g. Digg), social bookmarking (e.g. Delicious), social networks (e.g.
Facebook, MySpace and LinkedIn).
Definition
Modern datasets derived from telecommunication technologies such as online social
media and mobile phone systems offer great potential to understand the behaviors of
large populations during emergencies and disasters. This article reviews recent studies
using large-scale, modern data to understand emergency and disaster response, covering
work focused on social network activity during earthquakes and disease outbreaks and
mobile phone communications following bombing and other emergency events. The key
techniques and research trends are also discussed.
Page 3
3
Introduction
Large-scale emergencies and disasters are an ever-present threat to human society.
With growing populations and looming threat of global climate change, the numbers
of people at risk will continue to grow. Thus there is great need to optimize response
efforts from search and rescue to food and resource disbursement. Human dynamics
research offers a promising avenue to understand the behaviors of large populations, and
modern datasets derived from cutting-edge telecommunications such as online social
media and pervasive mobile phone systems bring a wealth of potential new information.
Such massive data offers a promising complement to existing research efforts in disaster
sociology, which primarily focus on eyewitness interviews, surveys and other in-depth
but small-scale data [18].
Yet most current human dynamics research is focused primarily on data col-
lected under normal circumstances, capturing baseline activity patterns. Here we re-
view a number of studies pushing the envelope of modern data into the realm of unex-
pected deviations in these population behaviors. We discuss research focused on mas-
sive datasets from social network activity during earthquakes and disease outbreaks to
mobile phone communications following bombings, power outages, and more.
We review a number of recent studies using large-scale, modern data to un-
derstand emergency and disaster response. We begin with a review quantifying how
expectations of communication in today’s world may influence our perception of the
severity of an emergency. We then cover works focused on social media and mobile
phones. These works use Twitter, a prominent online social media service, to under-
stand more about disease outbreaks and the impact of earthquakes. The mobile phone
studies feature a number of emergencies, including earthquakes, bombings, a plane
crash, and more. The results of these studies have the potential to revolutionize disas-
ter response in the future, with the critical goal of saving lives.
Page 4
4
Historical Background
Connectivity and information access through global telecommunications have become
increasingly pervasive due to modern technologies such as mobile phones and the inter-
net. People are becoming increasing reliant on these communication modes and so an
important question asked by Sheetz, et al [20] is: what do people expect about their ac-
cess to these communication channels when an emergency occurs? They explored how
the expectation of the availability of these communication technologies may influence
their perceptions of how they would use these technologies during and after a crisis.
To answer this question, the authors conducted online surveys and follow-up in-
terviews with Virginia Tech students, faculty, and staff (participants). This university
suffered a tragic attack on April 16, 2007 and the authors reported that local cellular
networks were overwhelmed by traffic. Surveying witnesses and survivors at the uni-
versity allows the authors to study how the perceptions of information access meshed
with the unfortunate events that occurred.
Through these surveys and interviews, they found that participants have a range
of expectations for connectedness in normal activities. Most participants did not expect
to be able to immediately contact someone. This held even for strong social ties, for
example a student trying to reach his or her parents. Most importantly, the authors
discovered that participants who do have high expectations of connectivity (and also
tend to be more extroverted individuals) were more likely to report problems with
connectivity than users with lesser expectations. These problems can lead these people
to form overestimate of the severity of the crisis, compared with individuals who have
lower expectations for their communication and are thus less likely to find communi-
cation loss a cause for concern. This means that an individual’s personal traits may
directly influence how he or she estimates the severity of a crisis.
Page 5
5
While the authors admitted that they had a small sample size and that their
interview methods may not be perfect, this study is an important step towards fur-
ther understanding the interplay between modern telecommunications and emergency
events.
Emergencies and Social Media
Today, social media such as Twitter and Facebook have been popularly used as every-
day communication tools. Millions of people use “tweets” or Facebook “statuses” to
inform family, friends, colleagues or any others about information, opinion and emo-
tions about events just happening, leading to the great potential of using social media
for monitoring and rescue purposes. Twitter allows users to send and receive tweets
(140-character messages) via text messages and internet-enabled devices, providing the
public with detailed anecdotal information about their surroundings. Given the real-
time nature of Twitter and the emerging social networking technologies, social media
has the potential to fundamentally alter our discussions of emergencies. We briefly re-
view some of the recent work on detecting disease outbreaks and earthquake response
with Twitter.
Twitter and disease outbreaks
Various studies have shown the potential of using Twitter data to monitor current the
public health status of a population, as people often tweet when they feel ill or recognize
disease symptoms. Quincey and Kostkova [17] collected tweets that contained instances
of the keyword “flu” in a week during the Swine Flu pandemic. Their study suggests
that the co-presence of other words in tweets can be used by public health authorities
to gather information regarding disease activity, early warning and infectious disease
outbreak. For example, in the majority of the collected tweets, the word “swine” was
Page 6
6
present along with “flu,” the words “have flu” and “has flu” may indicate that the tweet
contains information about the users or someone else having flu. The words “confirmed”
and “case(s)” perhaps indicate a number of tweets that are publicizing “confirmed
cases of swine flu”. Culotta [5] collected over 500,000 influenza-related tweets during 10
weeks and analyzed the correlation between these messages and the Centers for Disease
Control and Prevention (CDC) statistics. The paper reported a correlation of 0.78
by leveraging a document classifier. Chew and Eysenbach [4] collected over 2 million
tweets containing the keywords ‘H1N1’, ‘swine flu’ and ‘swinflu’ within eight months in
2009. Using manual and automated content coding, they found temporal correlation of
Twitter activity with major news stories and H1N1 incidence data. In addition, they
found that the majority of these tweets contained resource-related posts (e.g., links to
news websites). Gomide et al. [8] analyzed how the Dengue outbreaks in 2009 were
mentioned on Twitter. Using a linear regression model, they showed promising results
to predict the number of dengue cases by leveraging tweet content and spatiotemporal
information. Signorini et al. [21] tracked time-evolving public sentiments about H1N1 or
swine flu, and studied the probability of using Twitter stream for real-time estimation
of weekly influenza-like illness (ILI) statistics generated by CDC.
There has also been work addressing the technical challenges of collecting tweets
that are related to health or disease. Zamite et al. [22] proposed a system architecture
for collecting and integrating epidemiological data based on the principles of inter-
operability and modularity. Prier et al. [16] proposed using a Latent Dirichlet Allo-
cation (LDA) model to effectively identify health-related topics in Twitter. Paul and
Dredze [14] collected 2 billion tweets related to illness, diseases symptoms and treat-
ment from May 2009 to October 2010. They proposed a probabilistic aspect model to
separate tweets related to health from unrelated tweets. Aramaki et al. [1] collected 300
million tweets from 2008 to 2010. They applied the Support Vector Machines (SVMs)
Page 7
7
to find tweets related to influenza with a correlation 0.89% compared with Google Flu
Trend [7]. These tools offer the means to transform the overwhelming flood of big data
into more manageable information.
Besides social media, there are also other solutions to estimate a population’s
health from Internet activity, most notably Google’s Flu Trends service, which corre-
lates search term frequency with influenza statistics reported by the CDC [7].
Twitter and Earthquakes
In recent years, tremendous effort has been made towards leveraging Twitter to study
earthquakes, mainly falling into two lines of research: real-time detection [19, 9, 6] and
crisis management [10, 3, 12, 13].
Early earthquake detection and the delivery of timely alerts is an extremely chal-
lenging task. Depending on peculiarities of the earthquake, from size to location, alerts
may take between two and twenty minutes to publish, owing to the propagation time of
seismic energy from the epicenter to seismometers and the latencies in data collection
and validation. Therefore it has been practically impossible for affected populations
to know about an earthquake before it arrives. This situation is changing, however,
thanks to the pervasive use of Twitter. Users submit their tweets via text messages
and internet-enabled devices, and these messages are available to their followers and
the public within seconds, making Twitter an ideal environment for the dissemination
of breaking news to large populations. Therefore, by using populations as social sensors,
Twitter may be a viable tool for rapid assessment, reporting, and potentially real-time
detection of a hazard event. Sakaki et al. [19] investigated events such as earthquakes
and typhoons in Twitter, and proposed an algorithm to monitor tweets and to detect
earthquakes. They extracted features such as keywords in a tweet by semantic anal-
ysis, and used Support Vector Machines (SVMs) to classify a tweet into a positive
Page 8
8
or negative class. By regarding a tweet as a social sensor associated with location in-
formation, the authors transformed the earthquake detection problem into an object
detection problem in ubiquitous and pervasive computing. They derived a probabilistic
model by applying Kalman filtering and particle filtering to estimate the epicenter of
an earthquake and the trajectories of a typhoon. They then deployed an earthquake
reporting system in Japan, which delivers earthquake notifications to their users faster
than the announcements broadcast by Japan Meteorological Agency. Meanwhile, re-
searchers from the U.S. Geological Survey (USGS) reported an earthquake detection
system that adopts social network technologies, called Twitter Earthquake Detector
(TED) [9, 6]. They downloaded tweets that contain the words “earthquake”, “gempa”,
“temblor”, “terremoto”, or “sismo” from August to the end of November, 2009. Based
on tweet-frequency time series, they used a short-term-average, long-term-average al-
gorithm to identify earthquakes, finding 48 earthquakes around the globe with only
two false triggers in five months of data. The detections are faster than seismographic
detections, with 75% occurring within two minutes. These results demonstrate the ef-
ficiency of using Twitter as a detection tool, potentially achieving better and more
accurate results when combined with existing systems.
The rich semantics of tweets and Twitter’s broadcasting nature also hint at
the potential of using Twitter for rapid emergency response tools to assist in inter-
vention and crisis management. Caragea et al. developed a reusable information tech-
nology infrastructure, called Enhanced Messaging for the Emergency Response Sector
(EMERSE) [3]. The system is aimed at classifying tweets and text messages auto-
matically, together with the ability to deliver relevant information to relief workers.
EMERSE has four components, including an iPhone application, a Twitter crawler,
machine translation, and automatic message classification. The system analyzed the
information about the Haiti Earthquake relief, and provided their output to NGOs,
Page 9
9
relief workers, and victims and their friends and relatives in Haiti. To use Twitter as
an emergency response tool, it is important to assess the information quality of tweets
during an emergency situation. Li and Rao [12] studied Twitter usage following the
Sichuan Earthquake in China in 2008. They focused on five information quality di-
mensions: timeliness, accessibility, accuracy, completeness, and collective intelligence,
arguing that Twitter is an effective tool for information dissemination in critical mo-
ments following earthquake, and its broadcasting nature plays an important role in
emergency response. Mendoza et al. [13] studied the dissemination of false rumors and
confirmed news following 2010 Chile earthquake, finding that false rumors tend to be
questioned much more than confirmed news. Their study indicates the possibility of
using Twitter to detect rumors after an earthquake to make the rescue efforts more
efficient.
Emergencies and Mobile Phones
In addition to social media websites, the pervasive adoption of mobile phones provides
another, potentially even more detailed avenue to monitor large populations. Mobile
phone records usually include fine-grained longitudinal mobility traces and commu-
nication logs. The data allows greater opportunity to study personal social networks
through their relationship with physical space, compared to the online social networks
(e.g. “friends” and “followers” on Twitter). Mobile phones are well established in many
areas, even in third world countries such as Rwanda [11]. Leveraging their presence to
assist in emergency response has great potential to save lives. Here we review two re-
cent papers focused on mobile phones and emergencies. The first studied a earthquake
that occurred in central Africa [11]. The second analyzed a corpus of events including
non-emergency controls such as music festivals occurring in western Europe [2].
Page 10
10
An earthquake in central Africa
To understand how effective mobile phones are at understanding emergency situations,
a number of studies have been conducted. Kapoor, et al. [11] studied a 5.9 magnitude
earthquake that occurred Feb 3, 2008 in Lac Kivu region of the Democratic Republic
of Congo [11]. The dataset is the cellular activity patterns of mobile phone users in
Rwanda. They used daily call volume on a per tower basis, and they also had the
geographic coordinates of the towers. Their goal was to determine the location of the
epicenter algorithmically using only the cellular data, and to assess or predict what
areas of the country are most in need of aid due to the earthquake.
To study these problems, they assumed that (i) cell tower traffic deviates in a
statistically significant manner from normal activity levels and trends when an event
occurs; (ii) areas that are more disturbed by the event will display traffic deviations
for longer periods of time; (iii) disruptions are inversely proportional to the distance
from the catastrophe.
To detect an event they assumed the typical daily traffic on a tower obeys a
gaussian distribution and they used a negative log likelihood score to compare the
current traffic with this distribution. The higher this score, the more likely there was
an anomalous event on that day. They demonstrated that this score spikes on the day
of the event, although they did not discuss a specific algorithm to automatically flag
scores (for example, introducing a threshold score such that an event is anomalous
when its score exceeds that threshold).
To estimate the location of the event, they assumed the activity levels at a tower
during the event follow a normal or gaussian distribution but that the mean of this
tower’s distribution is now a function of the distance from the epicenter. Specifically
they used for tower i a distance-dependent mean mi + αDi(ex, ey)−1, where mi is the
normal mean traffic for i, α is some configurable scaling parameter, and Di(ex, ey) is
Page 11
11
the geographic distance of tower i from an epicenter located at coordinates (ex, ey).
They determined this epicenter (ex, ey) (and also α) using well-established maximum
likelihood estimates, i.e. they found the epicenter and scaling parameters that maximize
the sum of the log’s of all the tower’s probabilities.
The other problem they wish to address is to predict what areas are most in
need of emergency aid. To do this, they want to predict whether a particular tower will
experience a significant increase in traffic some number of days after the event. They
accomplished this by building a classifier which allows them to estimate this persistence
probability. Since it’s reasonable to assume that areas with higher populations are likely
to require more aid, they built an ”assistance opportunity score” for a location by taking
the product of the persistence probability estimate for that location and the population
at that location. Such a score allows emergency responders to potentially prioritize aid
efforts.
The authors also pointed out an important issue when using mobile phone data
to study these problems: the density of towers, and therefore information, is not uni-
form. Cities have many more towers than rural regions, and this leads to far greater
granularity in areas of high population, and greater information uncertainty in areas
with fewer towers. They exploited this fact to estimate what areas are most valuable to
survey manually for information after an event, by prioritizing surveys towards areas
with more uncertainty. They did this by devising a simple mechanism to drive down
the entropy in the information that may be gained from the system, and they even
incorporated geographic distances since it’s more expensive in terms of time and effort
to survey more remote regions.
All of their methods were validated by comparison with the Feb 3 earthquake
and were shown to work rather well. For future work they discussed a number of
interesting advancements such as incorporating richer models of geographic terrain.
Page 12
12
Mobile phones and disasters
Bagrow et al. [2] performed a data-driven analysis of a number of emergencies, including
bombings, a plane crash, another earthquake, and more. This work reported a number
of empirical discoveries regarding the response of populations in the wake of emergencies
(and non-emergency control events such as festivals), as measured from the country-
wide data of a single mobile phone provider in Western Europe. The assumptions made
by Kapoor et al. [11] are further justified by their work.
They found that emergencies trigger a sharp spike in call activity (number of
outgoing calls and text messages) in the physical proximity of the event, confirming that
mobile phones act as sensitive local “sociometers” to external societal perturbations.
In Fig. 1A, we plot the relative call volume ∆V/ 〈Vnormal〉 as a function of time, where
∆V = Vevent−〈Vnormal〉, Vevent is the number of calls made from nearby towers during the
event and 〈Vnormal〉 is the average call volume during the same time period of the week.
(Figure adapted from [2].) The anomalous traffic starts to decay immediately after the
emergency occurs, suggesting that the urge to communicate is strongest right at the
onset of the event. There was virtually no delay between the onset of the event and the
jump in call volume for events that were directly witnessed by the local population, such
as the bombing, the earthquake and the blackout. Brief delay was observed only for the
plane crash, which took place in an unpopulated area and thus lacked eyewitnesses. In
contrast, non-emergency events, like the festival and the concert, displayed a gradual
increase in call activity.
The temporally localized spikes in call activity (Fig. 1A) raises an important
question: is information about an event limited to the immediate vicinity of the emer-
gency or do emergencies, often immediately covered by national media, lead to spatially
extended changes in call activity [15]? To investigate this, Bagrow et al. inspected the
change in call activity in the vicinity of each event’s epicenter, finding that for the bomb-
Page 13
13
first half hour first hour first two hours Previousweek
C Bombing
G7
G0
A
B
BombingPlane crashEarthquakeBlackout
0
1
2
3
4
FestivalConcert
0 2 4 6 8
10
-3 -2 -1 0 1 2 3 4 5 6 7
ΔV/〈V
norm
al〉
time since event (hours)
-2 < t < 00 < t < 2
exp(- r / 9.98)
-2
0
2
4
6
8
10
0 10 20 30
× 10 3
distance from epicenter, r (km)
ΔV(
r) (c
umul
ativ
e)
Plane crash
Fig. 1: Temporal, spatial, and social response during emergencies. A, The time dependence of call
volume V (t) after four emergencies and two non-emergencies. We plot the relative change in call
volume ∆V/ 〈Vnormal〉, where ∆V = Vevent−〈Vnormal〉, Vevent is the call volume on the day of the event
and 〈Vnormal〉 is the average call volume during the same period of the week. B, The total change
in call volume during two-hour periods before and after the plane crash, as a function of distance
r from the epicenter of the crash. Following the event, we see an approximately exponential decay
∆V ∼ exp r/rc characterized by decay rate rc. C, Part of the contact network formed between mobile
phone users in the wake of the bombing. Nodes are colored by group, with G0 representing phone
users calling from the event region, G1 the recipients of those calls, etc. As time goes by more users
are contacted as information propagates. Those same users make little contact during a corresponding
time period the week before. (Figure adapted from [2].)
ing, for example, the change in call volume is strongest near the event, and drops rapidly
with the distance r from the epicenter. To quantify this effect across all emergencies,
they integrated the call volume over time in concentric shells of radius r centered on
the epicenter. The observed decay in anomalous traffic was approximately exponential,
∆V (r) ∼ exp (−r/rc), allowing one to characterize the spatial extent of the reaction
with a decay rate rc (We present their results for the plane crash in Fig. 1B). The
observed decay rates ranged from 2 km (bombing) to 10 km (plane crash), indicating
Page 14
14
that the anomalous call activity is limited to the event’s vicinity. An extended spatial
range (rc ≈ 110 km) was seen only for the earthquake. Meanwhile, non-emergencies
are highly localized: they possess decay rates less than 2 km. This systematic split
in rc between the spatially extended emergencies and well-localized non-emergencies
persisted for all explored events.
Despite the clear temporal and spatial localization of anomalous call activity
during emergencies, one expects some degree of information propagation beyond the
eyewitness population. To study how emergency information diffuses through a social
network, Bagrow et al. used mobile phone records to identify those individuals located
within the event region, forming a population called (G0), as well as a group called
G1 consisting of individuals outside the event region but who receive calls from the
G0 group during the event, a G2 group that receive calls from G1, and so on. They
reveal that the G0 individuals typically engage their social network within minutes, and
that the G1, G2, and occasionally even the G3 group show an anomalous call pattern
immediately after an emergency. We present their illustration of a segment of this
contact network for the bombing in Fig. 1C. The authors proceeded to further quantify
and control for this social propagation, and showed that the bombing and plane crash
have significant propagation up to the third and second neighbors of G0, respectively.
They found that other emergencies, the earthquake and blackout, displayed relatively
little propagation. This seems reasonable given the less severe nature of those events
(the earthquake was relatively minor).
Finally, we also presented a breakdown of a number of measurable features for
each emergency and non-emergency and showed that these features may be used to
distinguish anomalous call activity due to benign events such as music festivals from
spikes in call volume that indicate a dangerous event has occurred. Using such factors
may allow first responders to more accurately understand rapidly unfolding events, and
Page 15
15
may even allow them to actively solicit information from mobile phone users likely to
be near the event.
Key Techniques
We summarize the key techniques that have been used in the above-mentioned studies.
Event detection. The first challenge in large-scale emergency studies is to de-
termine and collect a subset of data relevant to emergencies under consideration. With
Twitter or other social media data where the communication content is available in
text format, most studies begin with a simple keyword matching, i.e., collecting data
that contained instances of the relevant keywords such as “flu”, “H1N1” and “earth-
quake.” The initial collections could be refined by manual and automated classification
process. Classification techniques such as Support Vector Machines (SVMs) have been
employed [1, 19], and topic clustering methods such as Latent Dirichlet Allocation
(LDA) can be used to improve the classification [14, 16]. Validation of this body of
work is often conducted based on authority reports such as Centers for Disease Control
and Prevention (CDC) statistics [21] (for disease outbreaks) or U.S. Geological Survey
(USGS) reports [9, 6] (for earthquakes). While the messages disseminated in social me-
dia might be inaccurate, there has been work on determining the quality of information
sources [12, 13]. Further, by applying time-series analysis and spatiotemporal pattern
analysis (e.g. Kalman filtering and particle filtering in [19]), researchers have developed
powerful earthquake detectors with performance comparative to existing earthquake
detection systems.
Event prediction and forecasting. The development of event prediction and
forecasting is still in its early stage. Gomide et al. [8] used a linear regression model to
predict the number of dengue cases. The earthquake detectors [19, 9, 6] that reported
earthquakes faster than the seismographic detection can be used as early warning
Page 16
16
system. There has been work on developing information infrastructure which has ability
to deliver relevant information to users once events are detected [3].
Spatiotemporal pattern recognition of events. Unlike social media data,
the content of communication is often unavailable in mobile phone data and hence
the identification of emergency events in mobile phone data relies analyses of spatial
and temporal anomalies of call logs. The main challenge of this research is to con-
struct reasonable null model in order to recognize anomaly events. Bagrow et al. [2]
proposed using pre-emergency normal activities as well as the activities during non-
emergency events to contrast the activities of emergency events. Based on this approach
the epicenter of an emergency event can be identified. Kapoor et al. [11] used a similar
methodology to identify event epicenters, as well as to predict the locations in need of
emergency aid.
Future Trends
Foundational work understanding the sociology of disaster was limited in scale by avail-
able data but surveys and interviews can ask a number of in-depth follow-up questions.
To understanding population response from, for example, mobile phone call volume
alone is potentially more challenging as such data, while perhaps being more objective,
is also far shallower. This begs the question: can more depth be found in communi-
cations data? The wealth of textual information available within social media such
as Twitter can be leveraged to learn more context about how populations respond to
emergencies, and advances in data mining and natural language processing techniques
offer the promise of even greater information. This may allow researchers to separate
relevant information from spurious activity, improving the accuracy and precision of
information available to rescuers.
Page 17
17
One can reasonably expect a degree of noise from any communication system, as
users will be focused on diverse topics. Yet when something of overwhelming importance
occurs, such as an emergency, it seems reasonable to expect that event to capture the
majority of user attention. This may lead to a communication system that is less noisy
and more focused as the severity of the event increases, in the sense that an increasing
fraction of the system’s communication will be about that event. Given this, it may
be worth trying to develop (rigorous) bounds on how much useful information can be
successfully extracted from such a system during and immediately following an event.
This could allow quantitative benchmarking of algorithms designed to assist rescuers
by comparing, for example, how much emergency information was extracted by an
algorithm with the maximum amount possible.
Meanwhile, it will be crucial going forward to develop algorithms that combine
and help understand multiple data sources—such as cell phone call volume, twitter mes-
sages and perhaps even security cameras, all from a given geographic locale. This trend
towards greater data availability and unification will only continue as more advanced
and entirely new forms of telecommunication come into widespread use. Without meth-
ods to handle the increased diversity and volume of communication, rescuers may be
unable to capitalize on the extra information provided by future telecommunications.
Conclusion
We have reviewed a number of works focused on the use of communication data, from
social media to mobile phones, to understand how people react to emergencies and dis-
asters. This problem is of critical importance: In many areas of the world, more people
than ever are at risk, as both human populations and threats due to climate change
continue to grow. Hopefully tools derived from social media and other communication
Page 18
18
datasets will help rescuers improve their emergency and disaster response by providing
accurate, useful, and timely information in the wake of such events.
Cross-References
1. Actionable Information in Social Networks, Diffusion of (00286)
2. Analysis Specific Social Media (00226)
3. Contextual Social Network Analysis (00229)
4. Counter-Terrorism, Social Network Analysis in (00281)
5. Disaster Response and Relief (00057)
6. Extracting Individual and Group Behavior from Mobility Data (00219)
7. Extracting Social Networks from Data (00011)
8. Impact of Social Networks on Human Behavior and Human Behavior in Social
Networks (00235)
9. Large Networks, Analysis of (00031)
10. Mechanisms for Information Propagation (00179)
11. Modeling and Analysis of Spatio-Temporal Social Networks (00320)
12. Models for Group Formation (00181)
13. Models of Social Networks (00182)
14. Social Media Mining and Knowledge Discovery (00188)
15. Social Network Datasets (00122)
16. Social Networking in the Telecom Industry (00343)
17. Spatio-Temporal Constraints on Social Networks (00321)
18. Spatio-Temporal Proximity and Social Distance (00326)
19. Temporal Networks (00042)
Page 19
19
References
1. E. Aramaki, S. Maskawa, and M. Morita. Twitter catches the flu: Detecting influenza epidemics
using twitter. In Proceedings of the Conference on Empirical Methods in Natural Language Pro-
cessing, pages 1568–1576. Association for Computational Linguistics, 2011.
2. J. P. Bagrow, D. Wang, and A.-L. Barabsi. Collective response of human populations to large-scale
emergencies. PLoS ONE, 6(3):e17680, 03 2011.
3. C. Caragea, N. McNeese, A. Jaiswal, G. Traylor, H. Kim, P. Mitra, D. Wu, A. Tapia, L. Giles,
B. Jansen, et al. Classifying text messages for the haiti earthquake. In Proceedings of the 8th
International ISCRAM Conference, ISCRAM, volume 11, 2011.
4. C. Chew and G. Eysenbach. Pandemics in the age of twitter: content analysis of tweets during
the 2009 h1n1 outbreak. PLoS One, 5(11):e14118, 2010.
5. A. Culotta. Towards detecting influenza epidemics by analyzing twitter messages. In Proceedings
of the First Workshop on Social Media Analytics, pages 115–122. ACM, 2010.
6. P. Earle, D. Bowden, and M. Guy. Twitter earthquake detection: earthquake monitoring in a
social world. Annals of Geophysics, 54(6), 2012.
7. J. Ginsberg, M. Mohebbi, R. Patel, L. Brammer, M. Smolinski, and L. Brilliant. Detecting
influenza epidemics using search engine query data. Nature, 457(7232):1012–1014, 2008.
8. J. Gomide, A. Veloso, W. Meira, V. Almeida, F. Benevenuto, F. Ferraz, and M. Teixeira. Dengue
surveillance based on a computational model of spatio-temporal locality of twitter. 2011.
9. M. Guy, P. Earle, C. Ostrum, K. Gruchalla, and S. Horvath. Integration and dissemination of
citizen reported and seismically derived earthquake information via social network technologies.
Advances in Intelligent Data Analysis IX, pages 42–53, 2010.
10. A. Hughes and L. Palen. Twitter adoption and use in mass convergence and emergency events.
International Journal of Emergency Management, 6(3):248–260, 2009.
11. A. Kapoor, N. Eagle, and E. Horvitz. People, quakes, and communications: Inferences from call
dynamics about a seismic event and its influences on a population. In Proceedings of AAAI
Artificial Intelligence for Development (AI-D’10), 2010.
12. J. Li and H. Rao. Twitter as a rapid response news service: An exploration in the context of the
2008 china earthquake. The Electronic Journal of Information Systems in Developing Countries,
42(0), 2010.
Page 20
20
13. M. Mendoza, B. Poblete, and C. Castillo. Twitter under crisis: Can we trust what we rt? In
Proceedings of the First Workshop on Social Media Analytics, pages 71–79. ACM, 2010.
14. M. Paul and M. Dredze. You are what you tweet: Analyzing twitter for public health. In Proceed-
ings of the 5th International AAAI Conference on Weblogs and Social Media (ICWSM), 2011.
15. M. Petrescu-Prahova and C. T. Butts. Emergent coordinators in the World Trade Center disaster.
International Journal of Mass Emergencies and Disasters, 28(3):133–168, 2008.
16. K. Prier, M. Smith, C. Giraud-Carrier, and C. Hanson. Identifying health-related topics on twitter.
Social Computing, Behavioral-Cultural Modeling and Prediction, pages 18–25, 2011.
17. E. Quincey and P. Kostkova. Early warning and outbreak detection using social networking
websites: The potential of twitter. Electronic Healthcare, pages 21–24, 2010.
18. H. Rodrıguez, E. Quarantelli, and R. Dynes. Handbook of disaster research. Springer Verlag, 2006.
19. T. Sakaki, M. Okazaki, and Y. Matsuo. Earthquake shakes twitter users: real-time event detection
by social sensors. In Proceedings of the 19th international conference on World wide web, pages
851–860. ACM, 2010.
20. S. Sheetz, A. L. Kavanaugh, F. Quek, B. J. Kim, and S.-C. Lu. The expectation of connectedness
and cell phone use in crises. Journal of Emergency Management, 2010.
21. A. Signorini, A. Segre, and P. Polgreen. The use of twitter to track levels of disease activity and
public concern in the us during the influenza a h1n1 pandemic. PLoS One, 6(5):e19467, 2011.
22. J. Zamite, F. Silva, F. Couto, and M. Silva. Medcollector: Multisource epidemic data collector.
Transactions on large-scale data-and knowledge-centered systems IV, pages 40–72, 2011.